I'm wondering if we shouldn't keep the linker consistent, in particular when linking both C and Fortran (but with a main in C).

I'm wondering if we shouldn't keep the linker consistent, in particular when linking both C and Fortran (but with a main in C).

This seems to work, at least for a simple test (is there an open-source C/Fortran project I should test?).

fortsub.f90

subroutine myfortsub (  ) bind ( C, name="myfortsub" )
   use mpi
   implicit none

    integer rank, size, ierror, tag, status(MPI_STATUS_SIZE)
    ! call MPI_INIT(ierror)
    call MPI_COMM_SIZE(MPI_COMM_WORLD, size, ierror)
    call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierror)
    print*, 'node', rank, ': Hello from Fortran'
    ! call MPI_FINALIZE(ierror)
   return
end subroutine myfortsub

main.c

#include <mpi.h>
#include <stdio.h>

extern void myfortsub();

int main() {
  MPI_Init(NULL, NULL);
  myfortsub();
  int size, rank, nameLen;
  char name[MPI_MAX_PROCESSOR_NAME];
  MPI_Comm_size(MPI_COMM_WORLD, &size);
  MPI_Comm_rank(MPI_COMM_WORLD, &rank);
  MPI_Get_processor_name(name, &nameLen);
  printf("[%d/%d] Hello from C! My name is %s.\n", rank, size, name);
  MPI_Finalize();
  return 0;
}

Build and link:

mpifort -c fortsub.f90
mpicc -c main.c
mpifort fortsub.o main.o -o hello
mpirun -np 8 ./hello

Output:

ld: warning: -commons use_dylibs is no longer supported, using error treatment instead
node           3 : Hello from Fortran
 node           2 : Hello from Fortran
[2/4] Hello from C! My name is Luke-MBP2021.fritz.box.
[3/4] Hello from C! My name is Luke-MBP2021.fritz.box.
 node           1 : Hello from Fortran
[1/4] Hello from C! My name is Luke-MBP2021.fritz.box.
 node           0 : Hello from Fortran
[0/4] Hello from C! My name is Luke-MBP2021.fritz.box.

The linker warning also occurs with the previous version of the forumula.

Sorry to invalidate the review. I forgot to bump the formula revision (which I think is necessary?).

Thanks for contributing to Homebrew! 🎉 It looks like you're having trouble with a CI failure. See our contribution guide for help. You may be most interested in the section on dealing with CI failures. You can find the CI logs in the Checks tab of your pull request.

Newbie here, so I don't really understand the procedures. I can locally reproduce the errors with opencoarrays and vineyard, and they are fixed when I brew install --build-from-source. Does this mean I should bump the revisions of these forumulae in my PR to get new bottles built?

The error with root test is the same (at least locally) whether I use the modified open-mpi formula from this PR or the existing one. My only guess is that the failure could be related to an Xcode update. Any ideas?

Does this mean I should bump the revisions of these forumulae in my PR to get new bottles built?

Yes for opencoarrays and vineyard. Ideally they'd also be fixed to not need a revision bump every time open-mpi is revision bumped but just revision bumping them is OK for now.

root is probably Xcode update so can ignore that one.

I'm afraid I don't understand the brew linkage errors. Are these indicating real issues caused by the changes to the open-mpi formula here?

@carlocab has requested bottles to be published to this PR.

⚠️ @carlocab bottle publish failed.

Looks like you need to rebase on master.

🤖 An automated task has requested bottles to be published to this PR.